Modelling and mapping canopy cover in African savanna using C-band Synthetic Aperture Radar (SAR); the case study of Bushbuckridge Local Municipality (BLM)

Khoza, Gladness Mikateko
Journal Title
Journal ISSN
Volume Title
The savanna biome covers approximately 20% of the earth’s land surface and more than half of the African continent. It is home to millions of people and different fauna and flora species but is under pressure due to natural and anthropogenic factors threating the existence of the biome. Remote sensing techniques have the capabilities of assessing savanna vegetation by mapping and monitoring savanna woody structure. Remote sensing systems are divided into passive(e.g., Landsat) and active sensors (e.g., Light Detection and Ranging-LiDAR and Synthetic Aperture Radar-SAR), with active sensors better suited for assessing woody structure than passive sensors. This is because their signals can penetrate through dense vegetation cover. LiDAR sensors are more accurate than SAR sensors, hence often used to train SAR-based models. However, LiDAR data is expensive to collect over large regions. Therefore, spaceborne SAR datasets are more useful for vegetation studies as they have a large spatial coverage and can capture large images at once. Due to the nature of savanna vegetation, monitoring vegetation within this biome is a challenging endeavour. This is because savannas are characterised by a mixture of grass and sparsely distributed trees mostly in the form of shrubs and woody vegetation. The success of monitoring savannas is dependent on the availability of up-to-date, accurate and well-validated spatial information. Accessing the information requires users to go through several pre-processing steps to turn the data into easily analysed formats. However, the pre-processing steps are challenging for users to implement and computationally expensive. Analysis Ready Data (ARD ) products promise a future where remote sensing data will no longer include computationally expensive or have challenging pre-processing steps but instead provides easy-to-use satellite data mapping and modelling savanna vegetation. This study assesses the utility of multi-temporal ‘Analysis Ready’ C-band SAR data to estimate woody canopy cover. It seeks to test different regression methods; Linear (LR), Support Vector Machine (SVM) and Random Forest (RF) regression models and their effectiveness in modelling woody canopy cover. Identify the optimal season for modelling canopy cover using different temporal combinations (dry & wet seasons of 2017 & 2018) and polarisations (VV & VH) of C-band SAR data, and subsequently mapping canopy cover using the most effective regression method and optimal season. The results show that although SVM produced higher results regardless of the season, noise or number of images that were analysed, it predicted for a smaller range while RF predicted for a larger range and performed better the more data or noise was added to the analysis. LR was not sensitive, or robust, enough to find good relationships between the variables as its performance dropped(compared to RF) as more images were added to the analysis. A combination of all seasons (dry & wet), years (2017 & 2018), and all polarisation bands (VV & VH) yielded higher results than when the images were analysed individually. Although the ARD products used in this study are still in the experimental phase, the results produced here are comparable to other savanna vegetation studies that used the operational C-band SAR data
A research report submitted to the Faculty of Science, University of the Witwatersrand, Johannesburg, in partial fulfilment of the requirements for the degree of Master of Science in GIS and Remote Sensing, 2021